Synapse

Monthly Newsletter from CNI

Issue - January 2024

Join us at the upcoming talk on "TTL Model for an LRU-Based Similarity Caching Policy" by Prof. Sala Alouf from Inria on 16/01/2024 at 18:00 hrs IST.

Join us at the upcoming talk on "Trade-offs between Security and Privacy in Linear Dynamical Systems " by Prof. Vaibhav Katewa, IISc Bangalore on 30/01/2024 at 17:00 hrs IST

Past CNI Scholar Anusree Rajan has defended her thesis titled “Resource-Aware State-Triggered Networked Control Systems.” Congratulations Dr. Anusree Rajan! Check out her publications here to learn more about her work.

Past CNI Scholar Krishna Chaythanya K.V. has defended his thesis titled “Generalized Likelihood Ratio Procedures for Early Detection of Machine Faults.” Congratulations Dr. Krishna Chaythanya!  

Sequential Clustering of Data Streams from Unknown Distributions

In this talk, Prof. Srikrishna Bhashyam addressed the problem of clustering data streams into K clusters with unknown underlying distributions. The proposed sequential nonparametric tests, for both known and unknown cluster numbers, show universal exponential consistency and finite-time convergence. Simulations indicate that the suggested sequential clustering tests require fewer samples than fixed-size tests to achieve a given error probability. 

Towards a Theory of Exploration in Continuous-Time Reinforcement Learning

Prof. Harsha Honnappa discusses applying reinforcement learning in continuous time to ultra-high-frequency interactions with stochastic environments. Focusing on realistic scenarios like stock market volatility modeled as fractional Brownian motion, the talk introduces a theoretical framework using rough path theory for exploring continuous-time stochastic environments. The proposed pathwise relaxed control framework models exploration processes in continuous time, emphasizing the establishment of the value function's existence and uniqueness. This work marks a foundational step towards a comprehensive theory of continuous-time reinforcement learning. 

CT Scan for Your Network: Topology Inference from End-to-end Measurements

In this presentation, Prof. Ting He discusses a technique known as network topology inference, focusing on obtaining network topology information through end-to-end measurements in the data plane. Unlike traditional methods that require access to the control plane or cooperation from internal nodes, this inference-based approach falls under the broader family of inversion problems called network tomography. The talk provides an overview of the basic idea, the current state of the art, limitations, and illustrates how the inferred information can enhance upper-layer applications within the context of overlay networks 

PhD Research: Elements of Excellence

Prof. Ness Shroff discussed strategies to succeed in a PhD program emphasizing honesty in research. He shared practical success tips, including the importance of thinking deeply about one’s problem, advisor selection, skill-building, and effective problem-solving. Drawing from thirty years of experience, he also gave advice on paper writing..